The transaction data should be immutable; many of the actual facts concerning the Banana are not mutable: it was picked when it was picked, it was moved when it was moved. Assuming the transactions reflect this accurately, they need not be mutable. What the transaction data should not be is irreversible. So, for example, if there's a data entry error discovered, I need to show that I'm changing the current knowledge of that banana without losing the prior erroneous transaction. It's the same sort of thing we'd do for a financial ledger: we don't change existing erroneous records in the database... we book new correcting records.

In a number of your comments you talk about errors: the transaction processing technology being discussed will not address those issues be they mistakes, non-compliance, or anything other that rather naive cheating.

I think people hear about some of the attributes of blockchain, such as being distributed, and say, "hey, that solves some of the problem of getting disparate, unrelated parties transacting in a common data set."... along with a misunderstanding of trust model and its relationship to data integrity across that shared ledger... and they get all excited about blockchain.

You're right you don't need blockchain and a central database could do it. The harder parts are administrative around the many parties and then getting the parties to participate at all. I think the transnational immutability aspects are beneficial for food safety, I don't think the raw complexity of implementing a new blockchain to do it makes sense.

I do think with the right kind of industry focused solution on top of something like this Amazon product, can be that sort of single database, with some of the helpful parts that have appeal in regards to blockchain and without the complexity. There are still issues which I don't believe something like this Amazon product solves, but I do think it moves the needle.